In 2023, large models represented by ChatGPT rapidly swept the globe, fostering an illusion of imminent technological leapfrogging. Despite industry perceptions that progress in large models over the past year surpassed that of the previous decade combined, the overall evolution of large model technology has entered a relatively plateaued stage.

With continually expanded and clarified cognition, enterprises have demystified large models, and an increasing number of them understand that large models are merely "technologies and capabilities." They are beginning to harness this new tool to enhance efficiency at the business level. This shift has led to a convergence in the trend of the "Hundred Models War."

Markets and customers are becoming increasingly pragmatic, with more startups shifting their focus from foundational models to applications and toolchains. For investors and investment institutions, the overarching concern remains commercialization.

However, there are still numerous challenges in realizing the commercialization of large models.

01

Challenge 1: ToB or toC

The first challenge lies in the vastly different and controversial approaches of large model players toward the to B and to C directions. Large models are indeed accelerating rapidly, but AI commercialization faces a significant gap. On one hand, traditional enterprises struggle to integrate AI into their original businesses, while on the other, AI enterprises find it difficult to identify suitable monetization methods. How can AI enterprises break through the dilemma of commercialization? The B-end and C-end present two distinct paths.

Looking at overseas AI players, many have forged their own commercialization paths. On the B-end, companies like Microsoft and Salesforce integrate AI technology into traditional products and offer vertical customization services. On the C-end, players such as OpenAI and Midjourney provide productivity-boosting tools to individual users and monetize through paid subscriptions.

While China started later, many enterprises are actively exploring commercialization paths, with companies like Baidu, Ali Group, ByteDance, 360, and iFLYTEK all engaging in relevant attempts. For instance, Baidu develops productivity tools for the C-end and introduces the Wenxin Yiyan subscription model, while also offering underlying architectures and solutions for the B-end. 360 leverages its browser's advantages in PC scenarios to focus on AI office applications for the C-end and AI security and knowledge management for the B-end, seeking commercial value across both ends. iFLYTEK, on the other hand, attempts to integrate large models with its hardware products.

Both the B-end and C-end face their respective challenges. On the B-end, traditional enterprises must consider ROI (return on investment), data security, and the high costs associated with integrating AI into existing workflows and subsequent maintenance. Based on the inertia in the enterprise service sector, large model applications for the B-end can bring greater industry value and can be realized faster, but traditional enterprises are generally hesitant to adopt them, with executives unwilling to use them and companies reluctant to pay for software.

On the C-end, while ordinary consumers' willingness to pay for AI products is rising, revenue struggles to cover the high costs of training and running large models. Furthermore, many enterprises tend to focus excessively on AI technology itself, neglecting the exploration of consumer markets and the mining of consumer needs.

Compared to C-end users who lack willingness to pay, B-end clients have a clearer demand for advanced technologies. IDC, the International Data Corporation, conducted an AI application survey in the fourth quarter of 2023, revealing that only 7% of the 100 surveyed enterprises had no plans for generative AI, indicating that over 90% of surveyed enterprises had already deployed AI applications. Among them, 24% had invested in generative AI with a clear budget, 34% had begun developing potential application scenarios, and 35% had started pilot projects but had not yet secured a clear budget.

However, opinions differ. For example, Li Kaifu, the founder of Sinovation Ventures and CEO of 01 Universe, believes that in the short term, large models have more opportunities for to C applications in China, but the challenge lies in high inference costs. When considering product-market fit, factors such as technical requirements, technical difficulty, and costs must also be taken into account, and the timing window must be seized.

For large model startups, finding a commercialization path, especially a business model in the to C arena, is a global conundrum.

Taking the popular Dark Side of the Moon as an example, for those who are resolute in betting on the to C path, constructing an effective business model in this arena is even more challenging. Yang Zhilin has also stated that currently, two popular business models exist: subscriptions and commissions. In his view, charging based on user numbers fails to generate significant commercial value as products evolve, and subscriptions are unlikely to be the ultimate business model.

The second is commissions. Advertising, which has been validated by the internet, offers higher certainty, but human attention and time are limited, limiting the opportunities for this business model.

Regardless of whether it's to B or to C, commercial performance is immediately apparent, prompting a rapid shift into a cutthroat competition phase. Investors or markets expect large model startups to introduce new products or undergo significant changes every 3-6 months, delivering tangible results, such as user numbers, revenue, and influence, that can convince customers or investors.

02

Challenge 2: Economics

Whether targeting the B-end or C-end, the core of large model commercialization ultimately boils down to an economic issue: whether revenue can cover costs.

"Large models are unaffordable to invest in, while small models lack profitability." An insider used this phrase to describe the current investment dilemma in the large model field.

On the one hand, the "expense" of large models is well-known, with training costs of tens of millions of yuan per run, ensuring that this is a game for a select few. Coupled with the end of the golden age of US dollar funds and increasingly cautious venture capital investments, it may be precisely for these reasons that the large model craze has struggled to ignite the primary market.

According to the "2023 AI Industry Status Report" released by research firm CB Insights, in 2023, China's AI sector saw approximately 232 investment and financing deals, a year-on-year decrease of 38%, with total financing amounting to approximately USD 2 billion, a year-on-year drop of 70%. Notably, the first quarter of 2023 marked the lowest levels of both funding volume and deal count in five years, indicating a pronounced "cooling" of AI financing in China.

As entrepreneurs face increasing difficulties in securing funding and covering the costs of long-term research, they must swiftly achieve commercial outcomes and complete business loops to ensure the sustainability of their projects – this has become the most significant difference between China's and Silicon Valley's large model startup ecosystems.

03

Challenge 3: Human Cognition

Apart from customer orientation and costs, human acceptance and usage of AI also pose challenges to the commercialization of large AI models.

At the 2024 World Artificial Intelligence Conference, Wang Jian, academician of the Chinese Academy of Engineering, Director of the Zhejiang Lab, and founder of Ali Cloud, stated, "The human factor is easily overlooked. When we say AI will impact every department and is a revolutionary force, getting everyone in all departments to embrace AI is difficult in many large enterprises. The difference between small and large enterprises is that large enterprises view AI as a revolutionary tool, while small enterprises see it as a tool for revolution."

Wang Jian observed that the current attitude of Chinese enterprises toward AI reflects the difficulty in getting all enterprises to embrace AI and large models. China Mobile began developing large models in early 2023 and launched a 13.9 billion-parameter large language model that year, accelerating the implementation of large models within the company and among clients. However, a significant challenge lies in how the industry views and embraces large models, with varying perceptions and acceptance levels within the industry, posing a challenge in mindset transformation.

This difficulty in mindset transformation exists not only within companies and industries but also within each of us.

From Socrates' assertion that "Man is the measure of all things" to Hamlet's "What a piece of work is a man! How noble in reason! How infinite in faculty! In form and moving how express and admirable!" and on to modern scientific theories laying the foundation for the Industrial Revolution, where productivity developments drove societal progress, humans have undoubtedly been the absolute protagonists of the past.

However, with the emergence of AI, many argue that technology will become the world's "protagonist," initiating a "battle of division of labor" between computers and humans – humans no longer have the exclusive privilege of producing knowledge, as machines can now do so too.

Against this backdrop, how humans can better divide labor with AI has become a topic of the times. While the answer may vary for each individual, it is undeniable that the AI era is approaching.

04

Conclusion

Patience with New Things

Ultimately, despite the various challenges in the commercialization of large AI models, the author believes that one day, large AI models will find suitable business models that work for them.

This confidence stems from the fact that AI represents advanced productivity. As Marx put it, new things are invincible because they represent the direction of development, conform to the laws of development, and are suitable for current and especially future development conditions, thus possessing strong vitality and broad prospects for development.

Regarding the current issues, we might as well adopt a dynamic and long-term perspective. After all, all problems must be resolved in a dynamic process – we cannot solve problems ten years from now based on the current state, but rather with the state ten years from now.

Therefore, we have reason to believe that the problems encountered in the commercialization of large models in China will inevitably be resolved through the dynamic development of large models.

- End -